On 2016/11/13 21:26, Nigel Taylor wrote: > > > I will not be able to test building that for sometime as in middle of > dpb run, I would say that it will fail. Not proved, but the estdc++ in > the wantlib requires the library to be installed, that requires a > dependency set by x11/qt5.port.mk, > > MODQT5_USE_GCC4_MODULE ?= Yes > .if ${MODQT5_USE_GCC4_MODULE} == "Yes" > MODULES += gcc4 > MODGCC4_LANGS += c++ > MODGCC4_ARCHS ?= * > .endif > > which means either qt5 and gcc4 plus the library is installed or the > library isn't installed then relies on the chance dpb has installed gcc4 > the library in building something else. Before stdc++ from base was used > for earlier version with qt4. > > This would require a LDEP change, and possibly a revision bump also, or > the inclusion of the gcc4 module and revision bump.
Yep, valid concern. As well as the ones you've suggested there's another option too. Any of these should work: 1. add the gcc4 module to doxygen so that the no_gui pseudo-flavour is also built with ports gcc. simple, but downside: pushes anything using doxygen later in dpb. 2. turn the no_gui pseudo-flavour back into a real flavour with a PFRAG for the gui bits. @pkgpath devel/doxygen,-gui would go in the pfrag. 3. split the gui off to a different port. keep the same pkgname stem, so it just needs an @pkgpath in PLIST for updates to work. Whichever is chosen, the current situation is wrong because no_gui is only a pseudo-flavour so it is only meant to select which subpackages are built and not change anything else; however at present it is also changing which compiler is used.